A dominant set-informed interpretable fuzzy system for automated diagnosis of dementia

Tianhua Chen, Pan Su, Yinghua Shen, Lu Chen, Mufti Mahmud, Yitian Zhao, Grigoris AntoniouFrontiers in Neuroscience 2022 · DOI

Abstract

Dementia is an incurable neurodegenerative disease primarily affecting the older population, for which the World Health Organisation has set to promoting early diagnosis and timely management as one of the primary goals for dementia care. While a range of popular machine learning algorithms and their variants have been applied for dementia diagnosis, fuzzy systems, which have been known effective in dealing with uncertainty and offer to explicitly reason how a diagnosis can be inferred, sporadically appear in recent literature. Given the advantages of a fuzzy rule-based model, which could potentially result in a clinical decision support system that offers understandable rules and a transparent inference process to support dementia diagnosis, this paper proposes a novel fuzzy inference system by adapting the concept of dominant sets that arise from the study of graph theory. A peeling-off strategy is used to iteratively extract from the constructed edge-weighted graph a collection of dominant sets. Each dominant set is further converted into a parameterized fuzzy rule, which is finally optimized in a supervised adaptive network-based fuzzy inference framework. An illustrative example is provided that demonstrates the interpretable rules and the transparent reasoning process of reaching a decision. Further systematic experiments conducted on data from the Open Access Series of Imaging Studies (OASIS) repository, also validate its superior performance over alternative methods.

Introduction

Dementia diagnosis is a critical race against time. While the World Health Organization emphasizes early diagnosis, the process is often manual, lengthy, and prone to practitioner bias. Machine learning (ML) offers a solution, but a significant barrier remains: the most accurate models, such as Deep Learning or Support Vector Machines, are often "black boxes." They provide a diagnosis without explaining the underlying reasoning.

In clinical settings, doctors require Explainable AI (XAI)—tools that offer a transparent reasoning process. Fuzzy systems are ideal for this because they use linguistic variables (e.g., "MMSE score is around 24") that mimic human reasoning. However, traditional fuzzy systems are difficult to construct, often requiring experts to manually define rules. This paper introduces DS-ANFIS, a novel system that automatically learns these rules using graph theory.

Methodology: The DS-ANFIS Pipeline

The researchers developed a four-stage pipeline to automate rule discovery and optimization:

  1. Graph Generation: The patient dataset is represented as a graph where nodes are patients and edges represent similarity based on clinical and MRI features.
  2. Dominant Set (DS) Clustering: Using a "peeling-off" strategy, the system extracts "dominant sets"—groups of highly similar patients. This method automatically determines the number of clusters and filters out noisy data points.
  3. Fuzzy Rule Initialization: Each cluster is converted into a human-readable fuzzy rule. For example: "If MMSE is around 0.96 AND nWBV is around 0.48, then the test is Negative."
  4. Optimization (ANFIS): These rules are refined in an Adaptive Neuro-Fuzzy Inference System, which fine-tunes the fuzzy sets and rule weights to maximize diagnostic accuracy.

Dataset and Feature Analysis

The study utilized the OASIS-2 (Open Access Series of Imaging Studies) repository, a longitudinal dataset of 150 subjects aged 60 to 96. Key features included the Mini-Mental State Examination (MMSE) score, Normalized Whole Brain Volume (nWBV), and Estimated Total Intracranial Volume (eTIV).

Figure 6: Correlation heatmap showing the relationships between clinical features and the Clinical Dementia Rating (CDR).

As shown in the correlation analysis, features like MMSE and nWBV show significant correlations with the Clinical Dementia Rating (CDR), justifying their inclusion in the fuzzy rule-based logic. The demographic distribution of the dataset also highlights the prevalence of dementia across different groups.

Figure 5: Distribution of dementia status across male (M) and female (F) subjects in the study population.

Key Results

The DS-ANFIS system demonstrated that high accuracy does not have to come at the cost of transparency:

  • High Accuracy: The system achieved an average classification accuracy of 81.7%.
  • Competitive Performance: It statistically outperformed Support Vector Machines (SVM) and Decision Trees (J48), performing on par with powerful but opaque Random Forests.
  • Interpretability: Unlike black-box models, DS-ANFIS provides a clear map of its logic, allowing clinicians to see exactly which clinical factors led to a specific diagnosis.
  • Efficiency: The system is computationally lightweight, averaging less than 0.3 seconds per diagnosis.

Conclusion and Future Work

DS-ANFIS represents a significant step toward trustworthy clinical decision support. By bypassing the need for manual rule definition through graph theory, the system provides a scalable way to build interpretable models. Future research will focus on multi-class classification to distinguish between different stages of dementia severity and validating the model on larger, more diverse global datasets.

Summary updated

Abstract

Dementia is an incurable neurodegenerative disease primarily affecting the older population, for which the World Health Organisation has set to promoting early diagnosis and timely management as one of the primary goals for dementia care. While a range of popular machine learning algorithms and their variants have been applied for dementia diagnosis, fuzzy systems, which have been known effective in dealing with uncertainty and offer to explicitly reason how a diagnosis can be inferred, sporadically appear in recent literature. Given the advantages of a fuzzy rule-based model, which could potentially result in a clinical decision support system that offers understandable rules and a transparent inference process to support dementia diagnosis, this paper proposes a novel fuzzy inference system by adapting the concept of dominant sets that arise from the study of graph theory. A peeling-off strategy is used to iteratively extract from the constructed edge-weighted graph a collection of dominant sets. Each dominant set is further converted into a parameterized fuzzy rule, which is finally optimized in a supervised adaptive network-based fuzzy inference framework. An illustrative example is provided that demonstrates the interpretable rules and the transparent reasoning process of reaching a decision. Further systematic experiments conducted on data from the Open Access Series of Imaging Studies (OASIS) repository, also validate its superior performance over alternative methods.
BibTeX
@inproceedings{chen2022dominant,
  author = {Tianhua Chen and Pan Su and Yinghua Shen and Lu Chen and Mufti Mahmud and Yitian Zhao and Grigoris Antoniou},
  title = {{A dominant set-informed interpretable fuzzy system for automated diagnosis of dementia}},
  booktitle = {Frontiers in Neuroscience},
  year = {2022},
  doi = {10.3389/fnins.2022.867664},
}
Figure